Prediction method of rule engine and deep learning combined high-speed railway tunnel shaft construction method

Abstract

The invention provides a rule engine and deep learning combined prediction method for a high-speed railway tunnel shaft construction method, and the method arranges and constructs a data sample library to support algorithm training. Based on the data set, the invention provides a rule engine and deep learning combined scheme, namely RI-ACNTPP, which is specially used for parameter design of the high-speed railway tunnel shaft construction method. The performance of the RI-ACNTPP model and models such as a GBDT (gradient boosting decision tree), an RF (random forest), an MLP (multilayer perceptron), an SVM (support vector machine) and the like is evaluated through a contrast experiment. The result shows that the RI-ACNTPP model not only reaches the highest level in prediction accuracy, but also can maintain higher prediction precision in the case of small sample data. The deep learning technology is applied to design of the high-speed railway tunnel construction method, the theoretical basis of tunnel intelligent design is enriched, and technical support is provided for design and review of the high-speed railway tunnel construction method in the future.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel body construction, and specifically to a prediction method for high-speed railway tunnel body construction methods that combines a rule engine with deep learning. Background Art

[0002] In the field of railway construction, intelligent tunnel construction has become a leading trend in industry development. The new era has given railway builders a crucial mission: researching and implementing intelligent tunnel construction technologies. Looking ahead, advanced technologies such as the Internet of Things, big data, and artificial intelligence will further integrate, promoting the research and application of a range of intelligent tunnel technologies, including intelligent identification of surrounding rock, intelligent design, intelligent equipment, intelligent construction, intelligent monitoring and quality control, and intelligent collaborative management platforms.

[0003] Determining the construction method is a crucial step in railway tunnel design. Traditionally, this design process requires professionals to analyze the characteristics of each section of railway based on their experience, a process that consumes significant manpower and material resources. Traditional design methods require both a long design cycle and the training of a large number of professional designers.

[0004] While machine learning has achieved some success in intelligent design, many challenges remain to be overcome. Specifically, given that the design of high-speed railway tunnel construction methods is inherently a complex nonlinear reasoning problem, existing research primarily relies on traditional machine learning methods. These methods have relatively simple model structures, struggle to handle complex nonlinear relationships, are poorly adaptable to changes in data distribution, and cannot automatically focus on key features. Furthermore, traditional machine learning techniques have limitations in addressing imbalanced datasets. Summary of the invention

[0005] In response to the shortcomings of existing technologies, the present invention aims to provide a prediction method for high-speed railway tunnel construction methods that combines a rule engine with deep learning to address the issues raised in the aforementioned background technology. This method improves the precision and efficiency of construction method design. It also pioneers a new approach to intelligent, automated design, enhances data understanding, improves model generalization and decision-making quality, and effectively addresses recognition bias caused by uneven data distribution, ensuring accurate and reliable predictions for key minority categories.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a prediction method for high-speed railway tunnel construction methods combining a rule engine with deep learning, comprising the following steps: S1. Determine the influencing factors of tunnel construction parameter design by analyzing tunnel design cases and summarize the high-speed railway tunnel construction design data set; S2. Preprocess the data. The preprocessing includes data resampling, missing value handling, and data normalization, aiming to improve data quality, enhance the model training effect, and ultimately support more accurate design decisions. S3. By sorting out the rules and specifications that need to be followed in the construction design, build a rule engine. First, comprehensively sort out all the specification requirements related to the tunnel body construction design, and convert these specifications into clear logical rule statements to build a detailed rule table. When new data is input into the model, the system automatically identifies the corresponding rule trigger conditions and quickly outputs the corresponding rule execution results. In the case of meeting multiple rules, the system will select the solution with the highest safety level as the final decision to ensure the safety and reliability of the construction. For data samples that cannot be directly analyzed using existing rules, a deep learning framework is established for supplementary prediction. This framework integrates the strong feature extraction ability of the convolutional neural network and introduces an attention mechanism to make the model focus on the part of the input data that is most relevant to the construction method, aiming to improve the understanding ability of the data. The Focal loss function is introduced. Build a deep learning model and complete training and saving. The deep learning model includes: a data encoding module, a model module, and a loss function module. S4. Build multiple machine learning models as comparison models for performance comparison and determine the algorithm model suitable for this problem. The machine learning models include gradient boosting decision tree, random forest, multi-layer perceptron, and support vector machine. Select accuracy, precision, recall, and F1 score as accuracy evaluation indicators. By comparing the results of these evaluation indicators, understand the performance characteristics of the various models built, and accordingly select the most suitable model for actual application.

[0007] Furthermore, in step S1, by analyzing the factors affecting the design of the construction method of the high-speed railway tunnel body, 19 factors affecting the tunnel construction method design are obtained, and they are used as feature factors to create an experimental data set. The feature factors include: whether there is bias pressure, whether there is outcrop, surrounding rock grade, maximum water inrush, normal water inrush, number of faults crossed, special geological type, whether it is a special geological influence zone, environmental grade, weathering grade, ground stress, joint development degree, risk factors, burial depth, relative distance from the portal, sample interval length, relative starting section number, relative ending section number, line name, and tunnel name.

[0008] Furthermore, when creating the database, first divide the entire tunnel into several small segments. The division standard is to group those with the same or similar above-mentioned features into one segment, and then respectively count the above-mentioned features of each segment of the tunnel. Finally, a tunnel body design database is formed.

[0009] Further, in step S2, first fill in the missing values of the established dataset. By comprehensively checking the dataset, determine which features have missing values and the degree of missingness. For continuous features with a relatively small missing ratio and relatively uniform distribution, use the mean of the feature to fill in the missing values. For categorical variables, use conditional probability or the most frequently occurring category based on the information of other variables to fill in the missing values.

[0010] Further, resample the data. Through statistical analysis of the prediction parameters, it is found that there are significant differences between data of different categories. Adjust the dataset by resampling to address the problem of sample imbalance.

[0011] Further, during the resampling process, the sampling is performed as a whole extraction according to the tunnel name, so that the number of samples in each category reaches a relatively balanced state. Finally, a dataset including multiple tunnels and cross-section records of multiple railway tunnels is constructed.

[0012] Further, in data preprocessing, for continuous data, a normalization method of normalizing separately according to each tunnel is adopted. The normalization function used is: ; where represents the minimum value of a certain continuous feature in this group of tunnels, the maximum value of a certain continuous feature in this group of tunnels, and maps an original value x to a value in the interval [0,1] through max - min normalization .

[0013] Further, in the data encoding module, AutoDis is used to encode continuous features, and entity embedding technology is used to process categorical features to improve the adaptability of the model to different feature types. In the model part, an ACmix attention module is introduced on the basis of the basic convolutional neural network, enabling the model to focus on the part of the input data that is most relevant to the construction method, to improve the understanding ability of the data, and to improve the generalization ability and decision-making quality of the model.

[0014] Further, in the loss function module, to address the problem of class imbalance, the Focal loss function is introduced to solve the class imbalance problem encountered during the training process. Add a modulation factor to the standard cross-entropy loss function , where it contains an adjustable parameter γ≥0. Therefore, the Focal Loss (FL) is defined as: .

[0015] Furthermore, in step S4, the accuracy is the ratio of the number of samples correctly classified by the model to the total number of samples; the precision measures the overall performance of the model in all samples; the precision is one of the commonly used evaluation indicators in classification tasks, and the precision is used to measure how many of the samples predicted by the model to be positive are truly positive samples; the recall is the ratio of the number of samples correctly predicted by the model to the number of all true positive samples, and the recall focuses on the model's coverage of all true positive samples; the F1 Score is the harmonic mean of precision and recall, and the range of the F1 Score is between 0 and 1. The higher the value, the better. The larger the accuracy, precision, recall and F1 Score values, the better the prediction effect. The optimal model is selected based on the above-mentioned accuracy evaluation index.

[0016] Beneficial effects of the present invention: 1. This prediction method innovatively combines deep learning with tunnel engineering design, proposing a rule-based engine and deep learning approach: the Rule-Integrated Attention-Based Convolutional Neural Network for Tunnel Body Parameter Prediction (RI-ACNTPP). By integrating existing engineering rules and specifications, this model not only improves prediction efficiency but also ensures the rationality and reliability of prediction results. The deep learning model framework incorporates the ACmix module, combining the feature extraction capabilities of convolutional neural networks (CNNs) with the dynamic focus on key features by the attention mechanism. This significantly enhances the model's ability to model complex nonlinear relationships, addressing the technical bottleneck of traditional models' inability to automatically identify key design factors.

[0017] 2. This invention uses the Focal Loss function to adjust sample weights, allowing the model to focus more on rare sample categories during training, effectively mitigating prediction bias caused by uneven data distribution. This model not only enriches the theory of intelligent tunnel design but also significantly improves the accuracy and efficiency of construction method design. It provides strong technical support for the design and review of future high-speed railway tunnel construction methods and opens up a new path for intelligent and automated design.

[0018] 3. The present invention incorporates design rules and model branches, and predicts data that are applicable to the rules through the established rule engine, ensuring the accuracy and interpretability of the prediction of this part of the data. At the same time, it can speed up the training speed and make the model more focused on data samples with unclear rules. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1Flowchart of the prediction method for the construction method of the high - speed railway tunnel body combining a rule engine and deep learning according to the present invention; Figure 2 Flowchart of the method for predicting construction parameters of the high - speed railway tunnel body using the deep learning method integrating rules in the embodiment of the present invention; Figure 3 RI - ACNTPP model architecture diagram in the embodiment of the present invention; Figure 4 Deep learning model structure diagram in the embodiment of the present invention; Figure 5 Deep model training loss diagram in the embodiment of the present invention; Figure 6 Deep learning model confusion matrix diagram in the embodiment of the present invention. Detailed implementation manners

[0020] To make the technical means, creative features, achieved purposes and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific implementation manners.

[0021] Please refer to Figures 1 to 6 , the present invention provides the following technical solutions: A prediction method for the construction method of the high - speed railway tunnel body combining a rule engine and deep learning. First, by sorting out rule specifications, a model branch is designed. For data applicable to rules, prediction is carried out through the built - in rule engine, ensuring the accuracy and interpretability of the prediction of this part of the data. A deep learning framework is established to handle samples without clear rules. This framework not only integrates the powerful feature extraction ability of the convolutional neural network (CNN), but also introduces an attention mechanism to make the model focus on the part of the input data most relevant to the construction method, thereby improving the ability to understand data, enhancing the generalization ability and decision - making quality of the model, especially in the face of uncertain data. In addition, to optimize the performance of the model on class - imbalanced data, the Focalloss loss function is introduced. This improvement enables the model to pay more attention to the few - sample classes, effectively solving the recognition deviation problem caused by uneven data distribution, and ensuring the prediction accuracy and reliability on the key minority classes. It mainly includes the following steps: Step 1: By analyzing tunnel body design cases, consulting relevant specifications, and consulting design experts, the influencing factors of tunnel body construction parameter design are determined. Combining the existing high - speed railway tunnel body design data, a high - speed railway tunnel construction design dataset is sorted out and summarized.

[0022] By analyzing the factors affecting the design of the construction method of the tunnel body of high-speed railways, 19 factors affecting the design of the tunnel construction method are obtained, and they are used as characteristic factors to create an experimental data set. The characteristic factors include: whether there is bias pressure, whether it outcrops, the surrounding rock grade, the maximum water inrush, the normal water inrush, the number of faults crossed, the special geological type, whether it is a special geological influence zone, the environmental grade, the weathering grade, the in-situ stress, the degree of joint development, the risk factors, the burial depth, the relative distance from the portal, the length of the sample interval, the relative number of segments from the starting point, the relative number of segments from the ending point, the line name, and the tunnel name. When creating the database, first, the whole tunnel is segmented into several small segments. The segmentation criterion is to group those with the same or similar above-mentioned characteristics into one segment, and then the above-mentioned characteristics of each segment of the tunnel are respectively counted, and finally, the tunnel body design database is formed.

[0023] Step 2: To ensure that the data in the tunnel body design sample library can meet the high-standard requirements of subsequent analysis and modeling, the data is preprocessed. These processes include but are not limited to data resampling, missing value processing, and data normalization, aiming to improve the data quality, enhance the model training effect, and ultimately support more accurate design decisions.

[0024] In this embodiment, the missing values in the established data set are filled. By comprehensively checking the data set, it is determined which features have missing values and the degree of missing. For continuous features with a relatively small missing ratio and relatively uniform distribution, the mean value of the feature is used for filling. For categorical variables, conditional probability or the most frequently occurring category is used for filling according to the information of other variables.

[0025] The data is resampled. Through the statistical analysis of the prediction parameters, it is found that there are significant differences between different categories of data. The data set is adjusted by the resampling method to solve the problem of sample imbalance. During the resampling process, the sampling is extracted as a whole according to the tunnel name, so that the number of samples in each category reaches a relatively balanced state, and finally a data set including multiple tunnels and multiple railway tunnel section records is constructed.

[0026] In the data preprocessing, for continuous data, the normalization is carried out separately according to the tunnel, and the normalization function used is: ; where represents the minimum value of a certain continuous feature in this group of tunnels, is the maximum value of a certain continuous feature in this group of tunnels, and an original value x is mapped to a value in the interval [0,1] through the maximum-minimum normalization . In this embodiment, minA represents the minimum value in this group of tunnel segments, and maxA represents the maximum value in this group of tunnel segments.

[0027] Step 3. In this embodiment, a rule engine is constructed by sorting out the rules and specifications that need to be followed in the construction design. First, all the specification requirements related to the tunnel body construction design are comprehensively sorted out, and these specifications are converted into clear logical rule statements. Based on this, a detailed rule table is further constructed. When new data is input into the model, the system can automatically identify the corresponding rule trigger conditions and quickly output the corresponding rule execution results. When multiple rules are met, the system will select the solution with the highest safety level as the final decision to ensure the safety and reliability of the construction.

[0028] For data samples that cannot be directly analyzed using existing rules, the present invention also specifically designs a deep learning model framework for supplementary prediction: Establishing a deep learning framework to handle samples that do not apply to the rules not only integrates the powerful feature extraction ability of the convolutional neural network (CNN), but also introduces an attention mechanism to make the model focus on the part of the input data that is most relevant to the construction method, thereby improving the understanding ability of the data, enhancing the generalization ability and decision-making quality of the model, especially in the face of uncertain data. In addition, in order to optimize the performance of the model on class-imbalanced data, the Focal loss function is introduced; a deep learning model is built, trained, and saved. The deep learning model includes: a data encoding module, a model module, and a loss function module. This improvement enables the model to pay more attention to the few-shot classes, effectively solves the recognition bias problem caused by uneven data distribution, and ensures the prediction accuracy and reliability on the key minority classes.

[0029] For the data encoding module, AutoDis is used to encode continuous features, and entity embedding technology is used to process categorical features to improve the adaptability of the model to different feature types. In the model part, the ACMix attention module is introduced on the basis of the convolutional neural network to make the model focus on the part of the input data that is most relevant to the construction method, for improving the understanding ability of the data, enhancing the generalization ability and decision-making quality of the model.

[0030] For the loss function module, the Focal loss function is introduced to address the problem of class imbalance encountered during the training process. A modulating factor is added to the standard cross-entropy loss function , which contains an adjustable parameter γ≥0. Therefore, the Focal Loss (FL) is defined as: .

[0031] Through this comprehensive method that combines rule engines and deep learning technologies, the accuracy and efficiency of predicting the construction parameters of high-speed railway tunnel bodies have been greatly improved, providing strong support for practical engineering applications.

[0032] In this embodiment, by systematically sorting out the "Code for Design of Railway Tunnels" (TB 10003-2016) and related technical standards, and combining expert consultations, the rules for the construction design of high-speed railway tunnel bodies have been clarified, as shown in the following table: The tunnel construction method should be determined through comprehensive research considering factors such as engineering geology, hydrogeology, lining type, tunnel depth, tunnel length, and environmental constraints. Table 1 is the selection table for construction methods of double-track tunnels in various grades of surrounding rocks under Type I mechanized matching conditions.

[0033] Table 1 Selection Table for Construction Methods of Double-Track Tunnels in Various Grades of Surrounding Rocks under Type I Mechanized Matching Conditions

[0034] Note: In the table, "●" indicates recommended adoption, "○" indicates available, and corresponding adjustments can be made according to actual situations during construction; the selection of construction methods should consider the influence of bedding and other factors, and appropriate strengthening should be carried out on the basis of the above table. On this basis, a design rule table has been sorted out, and a rule engine has been built to support the automation and standardization of the design process.

[0035] Step 4: Build multiple machine learning models as comparison models for performance comparison, and determine the algorithm model suitable for this problem. The machine learning models include gradient boosting decision trees, random forests, multi-layer perceptrons, and support vector machines. The accuracy rate, precision rate, recall rate, and F1 score are selected as accuracy evaluation indicators. By comparing the results of these evaluation indicators, understand the performance characteristics of the various models built, and accordingly select the most suitable model for actual application.

[0036] In addition, the accuracy rate is the ratio of the number of samples correctly classified by the model to the total number of samples; the precision rate measures the overall performance of the model among all samples; the precision rate is one of the commonly used evaluation indicators in classification tasks, and the precision rate is used to measure how many of the samples predicted as positive class samples by the model are truly positive class samples; the recall rate is the ratio of the number of samples correctly predicted as positive class samples by the model to the number of all true positive class samples, and the recall rate focuses on the coverage rate of the model for all true positive class samples; the F1 Score is the harmonic mean of the precision rate and the recall rate. The range of the F1 Score is between 0 and 1, and the higher the value, the better. The larger the values of the accuracy rate, precision rate, recall rate, and F1 Score, the better the prediction effect. According to the above accuracy evaluation index, select the optimal model.

[0037] In this example, to verify the prediction accuracy of the present invention, several commonly used methods were selected and compared with the above indicators. These commonly used methods include: GBDT (Gradient Boosted Decision Tree), RF (Random Forest), MLP (Multi-Layer Perceptron), and SVM (Support Vector Machine). The prediction accuracy of the present invention method is superior to all benchmark methods, demonstrating the superiority of the method, as shown in the following table: Table 2

[0038] According to the various accuracy evaluation indices obtained above, the more suitable model is the RI-ACNTPP model.

[0039] The optimal model obtained in step 3 is saved and imported into the data set to predict the construction method of the high-speed railway tunnel body and evaluate the accuracy of the evaluation indicators.

[0040] Based on the above method, this example integrates deep learning with tunnel engineering design to innovatively propose a tunnel body parameter prediction model (RI-ACNTPP) based on an attention mechanism and convolutional neural networks. This model not only enriches the theory of intelligent tunnel design but also significantly improves the accuracy and efficiency of construction method design. It provides strong technical support for the design and review of future high-speed railway tunnel construction methods and opens up a new path for intelligent and automated design.

[0041] The model uses AutoDis to encode continuous features and entity embedding to encode categorical features, providing the model with rich feature representation capabilities. It then uses CNN and ACmix modules for feature learning, employing Focal Loss as a loss function to address class imbalance in the training data. This helps designers accurately predict high-speed railway tunnel construction methods, thereby determining the appropriate approach for high-speed railway tunnel construction.

[0042] The basic principles, main features and advantages of the present invention are shown and described above. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention.

[0043] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A prediction method for the construction method of the tunnel body of high-speed railway tunnels combined with a rule engine and deep learning, characterized in that , including the following steps: S1. Determine the influencing factors of the tunnel body construction parameter design by analyzing the tunnel body design cases, and summarize the high-speed railway tunnel construction design data set; S2. Preprocess the data. The preprocessing includes data resampling, missing value processing, and data normalization, aiming to improve the data quality, enhance the model training effect, and ultimately support more accurate design decisions; S3. By sorting out the rules and specifications that need to be followed in the construction design, build a rule engine. First, comprehensively sort out all the specification requirements related to the tunnel body construction design, and convert these specifications into clear logical rule statements to build a detailed rule table. When new data is input into the model, the system automatically identifies the corresponding rule trigger conditions and quickly outputs the corresponding rule execution results. When multiple rules are met, the system will select the solution with the highest safety level as the final decision to ensure the safety and reliability of the construction; for data samples that cannot be directly analyzed using existing rules, a deep learning framework is established for supplementary prediction. This framework integrates the strong feature extraction ability of the convolutional neural network, and introduces an attention mechanism to make the model focus on the part of the input data that is most relevant to the construction method, to improve the understanding ability of the data, and introduces the Focal loss function; build a deep learning model, and complete training and saving. The deep learning model includes: a data encoding module, a model module, and a loss function module; S4. Build multiple machine learning models as comparison models for performance comparison, and determine the algorithm model suitable for this problem. The machine learning models include gradient boosting decision tree, random forest, multi-layer perceptron, and support vector machine, and accuracy, precision, recall, and F1 score are selected as the accuracy evaluation indicators. By comparing the results of these evaluation indicators, understand the performance characteristics of the various models built, and accordingly select the most suitable model for actual application.

2. The prediction method of the construction method for the tunnel body of high-speed railway combined with a rule engine and deep learning according to claim 1, characterized in that: In step S1, by analyzing the factors affecting the design of the construction method of the high-speed railway tunnel body, 19 factors affecting the tunnel construction method design are obtained, and they are used as feature factors to create an experimental data set. The feature factors include: whether there is bias pressure, whether there is outcrop, surrounding rock grade, maximum water inrush, normal water inrush, number of faults crossed, special geological type, whether it is a special geological influence zone, environmental grade, weathering grade, ground stress, joint development degree, risk factors, burial depth, relative distance from the portal, sample interval length, relative number of starting segments, relative number of ending segments, line name, and tunnel name.

3. The prediction method of a construction method for the tunnel body of a high-speed railway combined with a rule engine and deep learning according to claim 2, characterized in that: When creating the database, first divide the whole tunnel into several small segments. The division criterion is to divide those with the same or similar above-mentioned characteristics into one segment, and then respectively count the above-mentioned characteristics of each segment of the tunnel, and finally form a tunnel body design database.

4. The prediction method of the construction method for the tunnel body of high-speed railway tunnels combining a rule engine and deep learning according to claim 1, characterized in that: In step S2, first, fill in the missing values of the established dataset. By comprehensively checking the dataset, determine which features have missing values and the degree of missingness. For continuous features with a relatively small missing proportion and relatively uniform distribution, fill in the missing values with the mean of the feature. For categorical variables, use conditional probability or the most frequently occurring category to fill in the missing values according to the information of other variables.

5. The prediction method of the construction method for the tunnel body of high-speed railway tunnels combining a rule engine and deep learning according to claim 4, characterized in that: Resample the data. Through statistical analysis of the prediction parameters, it is found that there are significant differences between data of different categories. Adjust the dataset by resampling to address the problem of sample imbalance.

6. The prediction method of a construction method for the tunnel body of a high-speed railway combined with a rule engine and deep learning according to claim 5, characterized in that: During the resampling process, samples are extracted as a whole according to the tunnel name, so that the number of samples in each category reaches a relatively balanced state. Finally, a dataset including multiple tunnels and cross-section records of multiple railway tunnels is constructed.

7. The prediction method of a construction method for the tunnel body of a high-speed railway combined with a rule engine and deep learning according to claim 6, characterized in that: In data preprocessing, for continuous data, normalization is performed separately for each tunnel. The normalization function used is: ； wherein, represents the minimum value of a certain continuous feature in this group of tunnels, the maximum value of a certain continuous feature in this group of tunnels, and maps an original value x to a value in the interval [0, 1] through max-min standardization .

8. The prediction method of the construction method for the tunnel body of high-speed railway combined with a rule engine and deep learning according to claim 1, characterized in that: In the data encoding module, AutoDis is used to encode continuous features, and entity embedding technology is used to process categorical features to improve the adaptability of the model to different feature types. In the model part, an ACmix attention module is introduced on the basis of the basic convolutional neural network to enable the model to focus on the part of the input data that is most relevant to the construction method, to improve the understanding ability of the data, and to improve the generalization ability and decision-making quality of the model.

9. The prediction method of the construction method for the tunnel body of high-speed railway tunnels combining a rule engine and deep learning according to claim 8, characterized in that: In the loss function module, the Focal loss function is introduced to address the problem of class imbalance during the training process. In the standard cross-entropy loss function a modulating factor is added, which contains an adjustable parameter γ≥0. Therefore, the Focal Loss (FL) is defined as: 。 10. The prediction method of the construction method of the tunnel body of high-speed railway combined with a rule engine and deep learning according to claim 1, characterized in that: In step S4, the accuracy rate is the ratio of the number of samples correctly classified by the model to the total number of samples; the precision rate measures the overall performance of the model among all samples; the precision rate is one of the commonly used evaluation indicators in classification tasks, and the precision rate is used to measure how many of the samples predicted as positive classes by the model are truly positive class samples; the recall rate is the ratio of the number of samples correctly predicted as positive classes by the model to the number of all true positive class samples, and the recall rate focuses on the coverage rate of the model for all true positive class samples; the F1 Score is the harmonic mean of the precision rate and the recall rate. The range of the F1 Score is between 0 and 1, and the higher the value, the better. The higher the values of the accuracy rate, precision rate, recall rate, and F1 Score, the better the prediction effect. According to the above accuracy evaluation indicators, select the optimal model.

Images